Claim: Peter shot George.
Backing: WitnessWstates that Peter shot George.
Generalization: Witnesses normally tell the truth.
Subconclusion: WitnessWis telling the truth.
The possibility exists that a witness could lie. But unless there is some evi- dence indicating that the witness is not telling the truth, it is a reasonable inference to draw the subconclusion that witnessWis telling the truth. This subconclusion then backs the claim that Peter shot George. But the infer- ence in this case is comparable to the one in the Tweety case. It shifts a so-called probative weight onto the claim, but that weight can be removed
or lessened by new evidence that might come into the case. The impor- tance of the initial statement in evaluating the argument is thus revealed.
The context of use of the whole sequence of argumentation that follows the initial statement is that of an ongoing or open investigation. New evidence can come in at any point in the sequence, until the investigation is closed.
Defeasibility is a vital characteristic of argumentation where the context is that of an open investigation. Thus Verheij was right to link the defeasibility of the appeal to witness testimony to what he called its dialectical aspect.
Such an argument needs to be judged in the context of a given process in which there is doubt or disagreement about some central proposition that is at issue. There are two sides to the process, and each side has a so-called burden of proof. The outcome is in a balance with a weight of presumption on both sides. The function of a defeasible argument, such as an appeal to witness testimony, is to bring forward a small weight of evidence that can tilt the balance slightly to one side or the other. But the argument is open to defeat as new evidence comes in, until the process of investigation is finally closed.
Witness testimony is useful as a kind of argumentation under certain conditions. First, there is a certain situation or set of data or presumed facts that we as investigators need to know about. We do not have direct access to these data. For example, they may be in the past, and we cannot directly re-experience the past. But some living person may have had access to the past situation we are interested in, and may have observed it, or at least we may have reason to think that she has. Even if the person is not living or otherwise available, she may have recorded her impressions of what she saw.
In such a case, there is a possibility that we could come to know about this set of data indirectly, through this person we presume to have been a witness of it. But two key assumptions need to be made. The first is that we must assume that this person really was in a position to get access to these data directly at the time. Part of accepting that the witness is in a position to know is that we need to assume that she still has this information. She would need to remember it reasonably well. The second key assumption is that she can and will convey the information to us with reasonable truthfulness.
These two assumptions about appeal to witness testimony are, in turn, based on other even more fundamental assumptions. One is that the witness has properties of what is called an agent in multiagent systems in computing (Wooldridge and Jennings,1995). Agents can interact with other agents in dialogues. Agents can carry out actions. It will be necessary to view a witness as an agent of a certain kind. And it will be necessary to see the questioner of the witness as an agent, as well. These two agents need to be able to engage in orderly goal-directed conversations (dialogues) with each other. A witness, as an agent, must have the capability for knowing about presumed facts and must have the capability of relaying these facts to another agent. In other words, there must be a kind of communication between the witness as an
agent and the questioner who, as another agent, tries to get access to these facts through questioning the agent. Otherwise, appeal to witness testimony as a form of argument makes no sense at all.
Based on this analysis of its central structure, the appeal to witness opinion can be said to have the following basic form of argument. The variableW stands for an agent that is a witness. A witness is an agent that has incoming information about things it can perceive as facts or data and that can relay that information to another agent. The variableAstands for a statement (or proposition, which is taken to be an equivalent term). The generalization is a general rule that links the premises to the conclusion (Bex, Prakken, Reed, and Walton,2003). It is not an absolute, universal generalization, but is taken in the form of argument below as a defeasible conditional.
Argument from Witness Testimony
Position to Know Premise: WitnessW is in a position to know whether Ais true or not.
Truth-Telling Premise: WitnessW is telling the truth (asWknows it).
Statement Premise: WitnessW states thatAis true (false).
Generalization: If a witnessW is in a position to know whetherAis true or not, andW is telling the truth (asW knows it), andWstates thatAis true (false), thenAis true (false).
Conclusion: Therefore (defeasibly)Ais true (false).
The first three premises are joined together as a conjunction that appears as the antecedent of the conditional expressed in the warrant. The warrant functions as an additional premise. Thus the inner structure of appeal to witness opinion as a form of argument is that of the following defeasible modus ponens(DMP) type of inference.
If witnessWis in a position to know whetherAis true or not, andWis telling the truth (asWknows it), andWstates that any proposition is true, thenAis true.
WitnessWis telling the truth (asWknows it).
WitnessWstates thatAis true.
ThereforeAis true.
Since this inference has themodus ponensform, many might think that it is deductively valid. In traditional logic, it is the accepted conventional wisdom that all inferences having the form of modus ponens must be deductively valid. But the above inference is not deductively valid, according to the unconventional account presented here. It is a defeasible inference, because the first premise is a defeasible conditional. Thus it is a structurally correct form of inference that can be used to transfer a probative weight from the premises to the conclusion. But it is not deductively valid.
8.1.Strict and DefeasibleModus PonensArguments
An example of DMP can be found in the logic textbook Introduction to Logic(Copi and Cohen,1998, p. 363). Following the traditional logic text- book approach, they claim that the following argument is deductively valid because it has themodus ponensform.
If he has a good lawyer then he will be acquitted.
He has a good lawyer.
Therefore he will be acquitted.
Copi and Cohen translate the first premise from natural language into log- ical symbolism using the material conditional, defined as only false where the antecedent is true and the consequent false. On this definition, if both premises are true, and the argument is deductively valid, the conclusion must be true too. But is this particular argument well represented as being deductively valid? After all, in the real world, you can have a good lawyer, but still not be acquitted. It would seem to make more sense to translate this argument in a different way, not one that makes it deductively valid.
The usual approach in AI is to use a nonmonotonic logic to represent this kind of argument, as opposed to using classical deductive logic, a monotonic system of reasoning. A monotonic inference is one in which the conclusion drawn from the set of premises will be preserved as a conclusion even if the premises are supplemented by new information. The kind of reasoning illus- trated in the Tweety case is nonmonotonic, because new information that comes in, such as the information that Tweety is a penguin or Tweety is a baby bird, can defeat the old conclusion that Tweety flies. Once the new premises are added to the inference, the original conclusion must be retracted. To model this kind of reasoning we need a nonmonotonic logic. Horty (2001) has presented a survey of formal nonmonotonic reasoning systems.
Schum (1994) developed a theory of reasoning about evidence based on Bayesian probabilities, Wigmore’s theory of evidence, and Toulmin’s anal- ysis of inferences. An important feature of Toulmin’s model of reasoning is the concept of an inference warranted by a generalization that is sub- ject to exceptions and ancillary evidence that supports a conclusion. Schum offers examples of generalizations such as “The events reported by police officers testifying under oath usually have occurred” (Schum,1994, p. 87).
These kinds of generalizations can apply to the particulars of a case, gen- erating a conclusion by a process of inference in which new information can strengthen or weaken the inferential step from the premises to the conclusion.
Verheij (2001, p. 232) theorized that argumentation schemes of the kind typically used in law can be modeled using such defeasible generalizations (1999, p. 113). He based his theory on a distinction between two rule-based
forms of inference. The first one is the deductively validmodus formof argu- ment familiar from deductive logic. It can be called strict modus ponens (SMP), since the rule in the first premise is strict, meaning it admits of no exceptions. One counterexample defeats it.
StrictModus Ponens(SMP) As a rule, ifPthenQ P
ThereforeQ
SMP is monotonic, because it always remains valid, not matter how much new information enters into consideration. There is also another type of inference that is similar to SMP, except that exceptions to the rule are allowed, and will not defeat the inference itself, even though they may call for retraction of the conclusion, once the exception becomes known.
DefeasibleModus Ponens(DMP)
As a rule, ifPthenQ, but subject to exceptions P
No exceptions are known yet ThereforeQ
DMP is a nonmonotonic type of inference, because it can fail as new infor- mation comes in. If an exception to the rule becomes known, as applied to the case at issue, the conclusion may have to be retracted. The problem we now have is when to use which form of argument.
On Verheij’s model (2000, p. 5), in an instance in which only strict rules are involved, SMP can be applied, but DMP needs to be applied in an instance where both strict rules and rules not admitting of exceptions might possibly come into play. This is a policy of being on the safe side.
For example, if the given argument is based on a universal generalization about all triangles, without exception SMP can be applied. But DMP needs to be applied to a case where the generalization might be subject to qualifi- cations.11It is a view maintained by Verheij that many of the most common argumentation schemes found in arguments in law have the DMP form.
Argument from witness testimony is a case in point. It can be cast into doubt by asking critical questions.
11 Terminology remains unsettled. Some have the opinion that DMP is not really amodus ponens type of argument, and therefore should not properly be called amodus ponensof any sort.
Others may concede that DMP can be categorized as a kind ofmodus ponensinference, as long as the distinction between it and the deductive form SMP is carefully drawn.
9. Factors in Evaluating Witness Testimony
What kinds of critical questions would be appropriate for evaluating the form of argument called appeal to witness testimony? Of course, one of the premises of the appeal to witness testimony cited above is the assumption that the witness is telling the truth. Witnesses are examined in courts and make statements under an oath to tell the truth. As indicated by the truth- telling premise, there would be a general presumption that the witness is telling the truth. So that is one obvious critical question. This premise can be a source of default if critical questions can be asked that raise doubts about whether it holds in a case. In legal examples, an appeal to witness testimony can go through several phases. First it could appear reasonable, and then it could default, as a doubt was raised.
The sequence of argumentation in the case of Peter’s allegedly having shot George can be further expanded to illustrate complex argumentation as typically used in this kind of case. Suppose, for example, that evidence is introduced that shows that witnessWhad a good reason to lie. Then the sequence of argumentation might run as follows.
Argument 9
Initial Statement: It should be investigated whether Peter shot George.
Claim: Peter shot George.
Backing: WitnessWstates that Peter shot George.
Generalization: Witnesses normally tell the truth.
Subconclusion: WitnessWis telling the truth.
Defeater: WitnessWhas a reason for lying.
Although the defeasible generalization that witnesses normally tell the truth has been a reasonable premise to accept prior to this point, introduction of the defeater undercuts the support previously given to the claim by this gen- eralization. Thus the inference to the subconclusion is cancelled, removing the support for the claim that Peter shot George.
Similarly, questions can be raised on whether the position to know pre- mise holds in argument 9. For example, suppose it was a dark night, and the witness could not have seen the details of the crime in the way he claimed.
The major premise would default and the argument would fail. This factor is made evident in the Federal Rules of Evidence.12 On the basis of Rule 602, the objection “lack of personal knowledge” may be made in a trial. For example, if the examiner asks a witness, “Where was Mr. Jones at 8:00 p.m. on July 12?” the cross-examiner may object by saying, “Objection, no showing of
12 The Federal Rules of Evidence will turn out to be important in understanding witness testimony as a form of evidence in Anglo-American trials, and their role will be explained in Chapter 3, Section7.
personal knowledge” (Park, Leonard, and Goldberg,1998, p. 84). To avoid such an objection, the examining lawyer needs to lay a foundation for the question by establishing the personal knowledge of the witness before asking the question. Thus the position to know premise is a very important assump- tion in the presentation and questioning of witness testimony as evidence in the context of a trial. The statement premise is significant because the exact wording of what the witness says can be vitally important. Often the wording of a claim suggests conclusions by implicature or innuendo that the witness may deny, or least may not testify to, as revealed by questioning.
Anderson, Schum, and Twining (2005) have used the method of Wigmore charts to analyze evidence in legal cases. They view (p. 98) our adversarial system as a procedure that uses principles of logical reasoning to resolve disputed questions of fact. They see testimonial evidence as based on a number of factors. The first factor (p. 65) is that the witness is claiming personal knowledge of the occurrence of the event in question. The second is the requirement forbidding hearsay. The problem here is that if the wit- ness cannot be questioned directly, what he says may have no better status than rumor or gossip (p. 66). The third factor is that of inferences drawn from what the witness claims to observe. The fourth factor is the credibility assessment of a testimonial assertion. They consider credibility under three additional factors: observational sensitivity, objectivity, and veracity (p. 68).
9.1.Other Systems
advokateis a software system for assessing the credibility of witnesses in forensic and legal investigations (Bromby and Hall,2002). Theadvokate acronym refers to the following eight witness reliability factors to be exam- ined in a case (p. 148):
A: amount of time the witness observed the perpetrator.
D: distance from the witness to the perpetrator.
V: visibility conditions at the time.
O: whether the line of observation was impeded.
K: whether the perpetrator was known to the witness.
A: any reasons for remembering the event or the perpetrator.
T: time elapsed since the event.
E: errors in the description of the perpetrator.
According to Bromby and Hall (2002, p. 148), witness reliability inferences can be made from these factors, and what they call “defeating values” are findings in the above categories that provide evidence that a witness is unre- liable. Clearly these factors are important for evaluating witness testimony, and the question is how they can be fitted to the analysis above based on the argumentation scheme. The answer is that these factors, excluding E, which is already covered by the critical questions, relate specifically to the position to know premise. They relate to the evaluation of how strong or
weak the position to know premise is in any given case. Accordingly, the following list of seven critical questions can be subsumed as raising kinds of doubts about implicit assumptions of the position to know premise. All are expressed in terms of the witness observing an event, suggesting visual observation, but appropriate adjustments can be made for other forms of sensory information such as hearing.
PK1: What is the length of time since the witness observed the event?
PK2: How close was the witness to the event?
PK3: How favorable were the visibility conditions at the time?
PK4: Was the line of observation impeded?
PK5: Was the person observed known to the witness?
PK6: Were there reasons (other than PK5) for remembering what was observed?
PK7: How much time has elapsed since the event?
In addition to these critical questions, there are a number of other fac- tors identified by Schum (1994, p. 107) that need to be taken into account.
Schum classifies these under three headings: observational sensitivity, objec- tivity, and veracity. Under observational sensitivity there are such factors as sensory defects, general physical condition, conditions of observation, and quality of observation. Under objectivity there are such factors as expecta- tions, bias, and memory-related factors. Under veracity there are such factors as previous convictions for crimes of dishonesty, other misconduct related to dishonesty, and character evidence regarding honesty. Schum and Mor- ris (2007) distinguish between two primary categories of assessing witness testimony called competence and credibility. Under competence they cite four factors: “appropriate sources”, “in a position to observe”, “understand- ing of what was observed”, and “ability to communicate”. Under credibility they cite the three factors noted above: veracity, objectivity, and observational sensitivity. In their analysis, they emphasize that these two major categories are frequently confused, leading to serious inferential errors. This method of classification is based on experience regarding witness testimony accumu- lated in our legal system since the year 1352 (Schum,1994, p. 106). Some of these factors appear similar to the factors cited byadvokate, while others suggest additional critical questions. It would seem then that in addition to the critical questions we have cited above, others need to be considered as well.
In the study of argumentation schemes, at present it is still an unsolved problem how to manage critical questions, in several respects. One prob- lem is whether the critical questions correspond to unstated assumptions in the premises of the scheme. Another is whether there can be critical subquestions under the main critical questions. Another is how many crit- ical questions have to be answered before an argument fitting a scheme is completely proved so that it can no longer be defeated as more information